14:25 orbea: iive: you left before I responded earlier, but yes, nodcc entirely fixes that game.
14:31 iive: orbea, great
14:32 iive: I hope you have reported the bug to the radeon people... even if it is a bit hard to reproduce :|
14:35 orbea: should I do it to the gitlab mesa tracker or somewhere else?
15:25 iive: i guess gitlab. maybe mannerov could give some hints or look into it.
15:26 iive: developers not been able to easily reproduce the problem is the biggest obstacle.
15:27 orbea: yea, I wonder why apitrace fails to capture it
15:27 iive: can you try an older mesa version? It would be extremely useful if it is regression and a bisect could find a specific commit.
15:28 orbea: possibly, but given llvm I likely can't go back far enough
15:28 iive: as I said, threading might be involved, making the bug random. mesa gallium has some internal threading...
15:29 orbea: with no way to disable that?
15:29 iive: gl_thread is likely opengl only, but there is at least one thread that feeds the buffers to the kernel driver...
15:31 iive: LP_NUM_THREADS=0 is about llvm, likely not related, but try it anyway.
15:32 iive: gah, can't find it in the documentation.
15:33 orbea: yea, LP_NUM_THREADS=0 doesn't help
15:37 iive: RADEON_THREAD , but I don't know if it works with amdgpu.
15:38 iive: i don't think that radeon.ko support vega.
15:38 orbea: yea I am using the amdgpu kernel driver
15:39 iive: try setting it to 0 anyway...
15:40 orbea: no luck
15:46 iive: maybe try some more AMD_DEBUG options, maybe some dma stuff might also fix it.
15:50 orbea: no luck with nodma, trying others now
15:51 orbea: notiling seems to fix it too
15:58 mannerov: there are gallium traces
15:58 mannerov: no idea how to make them
15:58 mannerov: but it exists
16:02 orbea: i dont see anything with a search yet
16:03 orbea: nowc causes the game to crash, but I guess that is not related
16:04 orbea: forcedma completely breaks the graphics and does not fix the issue
16:09 mannerov: Open a bug report on gitlab
16:10 mannerov: you'll see if they need more
16:10 orbea: will do in a little bit
16:53 orbea: made an issue but
16:53 orbea: bad paste
16:53 orbea: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2991
16:53 orbea: hope I didn't miss anything
19:12 mannerov: orbea: csmt_force=0 ?
19:12 orbea: mannerov: tried earlier, no luck
19:12 orbea:updates the issue to reflect that
19:13 mannerov: you should set radeonsi flag
19:13 orbea: ?
19:14 orbea: in the issue title?
19:14 orbea: idk if nouveau or iris is affected
19:18 mannerov: when you create gitlab issue you can assign flags
19:18 mannerov: the thing is in the way you created the bug report radeonsi devs won't look at it
19:18 mannerov: based on the title
19:19 mannerov: see the flags: https://gitlab.freedesktop.org/mesa/mesa/-/issues
19:19 orbea: you mean these tags? https://gitlab.freedesktop.org/mesa/mesa/-/issues?label_name%5B%5D=radeonsi
19:19 orbea: i dont think I can do that, don't have the permissions
19:21 mannerov: done
19:21 orbea: thanks :)
19:22 orbea: actually just checked, llvmpipe is affected too
19:23 mannerov: and you say apitrace cannot show that ?
19:23 orbea: yep, it replays fine with no bug
19:23 mannerov: are you sure llvmpipe is used ? Like it is super slow
19:24 mannerov: Does the game feature some MSAA
19:24 mannerov: tried thread_submit=true ?
19:24 orbea: from glxinfo: OpenGL renderer string: llvmpipe (LLVM 10.0.0, 256 bits)
19:24 orbea: i think so?
19:24 orbea: i dont notice any slowdown, but maybe the game is just old enough that it runs fine
19:25 mannerov: D3D_ALWAYS_SOFTWARE=1
19:25 mannerov: for llvmpipe
19:25 orbea: i guess I wasn't doing it right, it crashed with the llvmpipe...
19:27 orbea: with D3D_ALWAYS_SOFTWARE http://dpaste.com/1E8N82Y.txt
19:27 mannerov: it often crashes because MSAA
19:27 mannerov: llvmpipe cannot do it yet
19:28 orbea: msaa is disabled in the game config, but that might not be enough
21:42 iive: orbea, were you using wined3d to record the apitrace ?
22:26 orbea: iive: yes
22:27 orbea: i recorded with wined3d and replayed with nine
22:36 iive: orbea, have you tried recording with nine?
22:36 iive: in case nine uses some different format or something.
22:41 orbea: i'll try
22:59 orbea: iive: the trace with nine replays as a black screen :\
22:59 orbea: also tried making sure apitrace was fully updated, no different (except for being much slower...)
23:09 orbea: the issue does appear when recording a trace with nine though, but replaying is just black so that is not useful...